Shrinking ponds in subarctic Alaska based on 1950–2002 remotely sensed images



[1] Over the past 50 years, Alaska has experienced a warming climate with longer growing seasons, increased potential evapotranspiration, and permafrost warming. Research from the Seward Peninsula and Kenai Peninsula has demonstrated a substantial landscape-level trend in the reduction of surface water and number of closed-basin ponds. We investigated whether this drying trend occurred at nine other regions throughout Alaska. One study region was from the Arctic Coastal Plain where deep permafrost occurs continuously across the landscape. The other eight study regions were from the boreal forest regions where discontinuous permafrost occurs. Mean annual precipitation across the study regions ranged from 100 to over 700 mm yr−1. We used remotely sensed imagery from the 1950s to 2002 to inventory over 10,000 closed-basin ponds from at least three periods from this time span. We found a reduction in the area and number of shallow, closed-basin ponds for all boreal regions. In contrast, the Arctic Coastal Plain region had negligible change in the area of closed-basin ponds. Since the 1950s, surface water area of closed-basin ponds included in this analysis decreased by 31 to 4 percent, and the total number of closed-basin ponds surveyed within each study region decreased from 54 to 5 percent. There was a significant increasing trend in annual mean temperature and potential evapotranspiration since the 1950s for all study regions. There was no significant trend in annual precipitation during the same period. The regional trend of shrinking ponds may be due to increased drainage as permafrost warms, or increased evapotranspiration during a warmer and extended growing season.

1. Introduction

[2] Recent environmental changes associated with climate warming at high latitudes have been summarized [Serreze et al., 2000; Overland et al., 2002, 2004; Hinzman et al., 2005] and include abiotic changes such as reduction in annual snow cover [Brown and Goodison, 1996; Brown and Braaten, 1998], reduction in sea ice extent [Bjørgo et al., 1997; Cavalieri et al., 1997], warming and thawing of permafrost [Osterkamp and Romanovsky, 1999; Camill, 2005], increased thermokarsting [Jorgenson et al., 2001], warming of lake water [Hobbie et al., 2003], and earlier spring river and lake ice break-up [Keyser et al., 2000; Magnuson et al., 2000]. Biotic changes at high latitudes have also occurred including expansion of boreal forest into tundra [Suarez et al., 1999], increased shrubbiness in tundra [Chapin et al., 1995; Sturm et al., 2001], expansion of shrubs into drying lake beds [Klein et al., 2005], and increased growing season and primary production [Zhou et al., 2001; Stow et al., 2003].

[3] In subarctic interior Alaska, the growing season climate regime switched from a predominantly cool and moist to hot and dry after a Pacific-wide regime shift in 1977 [Mantua et al., 1997; Hare et al., 1999; Hare and Mantua, 2000; Dickson, 2000]. On the basis of tree-ring reconstructed climate and recorded summer temperature observations, some of the warmest summers over the past 200 years in interior Alaska have occurred since the 1970s, while the coolest growing seasons occurred in the early to mid 1900s [Barber et al., 2004]. On the basis of tree-ring analysis, drought stress associated with a warming climate in Alaska may be widespread at nontreeline sites [Barber et al., 2000; Lloyd and Fastie, 2002; Wilmking et al., 2004].

[4] Climate warming may cause a significant change in regional hydrology in subarctic regions. For example, Roulet et al. [1992] estimated a drop in fen water table level of 190 to 240 mm under a climate warming scenario. While precipitation in interior Alaska has not changed substantially in recent decades, warming temperatures have led to increases in potential evapotranspiration (PET) for Alaska tundra regions [Oechel et al., 2000]. In this paper, we use remotely sensed imagery from the past 50 years to investigate whether there was a landscape-level change in closed basin ponds associated with the recent climate warming in Alaska.

2. Methods

2.1. Study Areas

[5] Our main study area is the subarctic boreal region of interior Alaska. This region spans over 5 million km2, bounded on the north by the Brooks Range and on the south by the Alaska Range. These mountain ranges act as barriers to marine air masses and consequently Interior Alaska is semi-arid with an annual precipitation of 500 to <200 mm [Fleming et al., 2000; Hammond and Yarie, 1996]. Interior Alaska has experienced an increase in growing season length [Keyser et al., 2000] and a warmer, drier climate [Barber et al., 2000, 2004]. It is also a region of discontinuous permafrost that is warming and degrading [Osterkamp and Romanovsky, 1999; Jorgenson et al., 2001] which may lead to drier soils in some boreal sites [Camill, 2005].

[6] Our selection of study regions in interior Alaska was constrained by the availability of cloud-free aerial photography and satellite imagery. Primarily on the basis of imagery availability, we selected low-lying tectonic basins (Figure 1) with imagery from at least three time periods ranging from the 1950s to 2002. To contrast our study regions from interior Alaska, we also selected a study region from the Arctic Coastal Plain, an area with a much colder and shorter growing season and underlain by continuous permafrost of hundreds of meters. We also selected one region (Talkeetna) from the Cook Inlet Basin, south of the Alaska Range. Talkeetna, which lies in a more maritime climate, had approximately twice the mean annual precipitation relative to interior Alaska study regions and was the only study region with a mean annual temperature above 0°C (Table 1). The Copper River Basin is south of the Alaska range, but has a continental climate similar to interior Alaska because it is in a rain shadow of the surrounding Chugach, Wrangell and Alaska mountain ranges. All study regions were alluvial or lacustrine plains with a maximum slope gradient of less than 5 percent.

Figure 1.

Alaska study regions and long-term first order weather stations used in the study. Study regions: 1, Arctic Coastal Plain; 2, Steven Village Area; 3, Yukon Flats National Wildlife Refuge; 4, Minto Flats State Game Refuge; 5, Denali National Park; 6, Talkeetna Area; 7, Innoko National Wildlife Refuge; 8, Tetlin National Wildlife Refuge; 9, Copper River Basin, Wrangell St. Elias National Park. Weather Stations: Ba, Barrow; Be, Bettles; Fa, Fairbanks; Gu, Gulkana; Mc, McGrath; No, Northway; Ta, Talkeetna.

Table 1. Study Regions Climatology and Imagery Datesa
Study RegionMean Annual Temperature 1971–2000, °CMean Precipitation 1971–2000, mmWeather StationStudy Region Area (×1000 ha)Number of Water Bodies DelineatedImagery Analyzed
Arctic Coastal Plain Region
Arctic Coastal Plain−11.8102Prudhoe Bay2614261954 aerial photography
      1978 aerial photography
      3 Aug 1999 Landsat ETM+
Interior Alaska Regions
Tetlin National Wildlife Refuge−4.9238Northway Airport9521521954 aerial photography: 22 August
      1978 aerial photography: 1 July
      1981 aerial photography: 1 August
      1999 Landsat ETM+: 4 August
Yukon Flats National Wildlife Refuge−3.9193Circle20213701952 aerial photography: 14 June,24 June, 5 July, 18 July, 21 August,
      13 September
      1979 orthoquad maps
      2000 Landsat ETM+: 16 August
Stevens Village−3.9193Circle10712181951 aerial photography: 27 June, 11 July
      1978 aerial photography: 1 July
      1991 Landsat TM: 29 June
      2000 Landsat ETM+: 16 August
      2002 Landsat ETM+: 6 August
Minto Flats State Game Refuge−2.9262Fairbanks Airport8113901951 aerial photography: 27 June, 11 July
      1954 aerial photography: 7 August
      1980 aerial photography: 1 July
      1991 Landsat TM: 29 June
      2000 Landsat ETM+: 16 August
      2002 Landsat ETM+: 6 August
Copper River Basin−2.7283Gulkana Airport301011955 aerial photography: 5 July
      1957 aerial photography: 29 July
      1978 aerial photography: 28 August
      1995 Landsat TM: 8 August
      2001 Landsat ETM+: 7 August
Denali National Park−2.8380McKinley Park828761951 aerial photography: 28 August
      1952 aerial photography: 20 July, 5 August
      1954 aerial photography: 7 August
      1979 aerial photography: 1 August
      1981 aerial photography: 23 August
      2000 Landsat ETM+: 16 August
Innoko Flats National Wildlife Refuge−2.8445McGrath Airport6312051952 aerial photography: 14 June, 24 June
      1980 aerial photography: 1 July
      2001 Landsat ETM+: 18 August
Cook Inlet Basin Region
Talkeetna1.1716Talkeetna Airport17710521951 aerial photography: 3 July
      1953 aerial photography: 3 July, 11 July
      1954 aerial photography: 17 June
      1980 aerial photography: 1 August
      2000 Landsat ETM+: 16 August

2.2. Remote Sensing Methods

[7] To estimate surface water changes, we used remotely sensed imagery from the 1950s to 2002. For each study region, we analyzed scanned panchromatic aerial photographs from the 1950s, scanned color infrared aerial photographs from 1978–1982, and digital images from the Landsat Enhanced Thematic Mapper Plus sensor (ETM+) from 1999–2002. Landsat Thematic Mapper TM images from 1991–1995 provided an extra observation period for some study regions (Table 1). Smoke from wildfires and clouds were a major limitation in acquiring useful satellite imagery from the summer months. Because of the high latitude, cloud shadows cover a greater area than clouds; for example an image with 25 percent cloud cover would be over 50 percent unusable owing to cloud and associated cloud shadows covering the Earth's surface.

[8] The 1950s panchromatic aerial photographs were paper prints at an original scale of 1:40,000 obtained from the U. S. Geological Survey (USGS), Anchorage, Alaska. The 1978–1982 aerial photographs were 1:60,000 in color infrared transparencies from the Alaska High Altitude Aerial Photography (AHAP) Program and available from the GeoData Center, University of Alaska Fairbanks. All photographs were scanned at 600 dpi to create digital images. Panchromatic digital ortho photographs were available for the Yukon Flats study area as a product from the AHAP program and were obtained from the Yukon Flats National Wildlife Refuge headquarters, Fairbanks, Alaska. Landsat TM and ETM+ digital images were ordered from the EROS Data Center, Sioux Falls, South Dakota. All images were from the period of mid-June to mid-August.

[9] All digital images were georectified to the UTM map projection, NAD27. We used the statewide coverage of 1:63,360 topographic maps as the source for control points. Each digital image was rectified using a second-order polynomial model based on at least 25 control points. Each rectification model had a maximum Root Mean Squared Error of less than 1 satellite image pixel (30 m). After completing the rectification process, we had a time series of spatially aligned images for each study region.

[10] We used closed ponds (lacking inlets or outlets) because they are sensitive to changes in climate in semiarid regions such as interior Alaska [Barber and Finney, 2000]. Initially we attempted automated classification of closed ponds. We abandoned automated spectral approaches commonly used in digital image processing due to problems from (1) sunglint near the edge of aerial photographs creating bright ponds that were confused with other bright targets such as meadows, and (2) cloud shadows creating dark patches that were spectrally similar to water.

[11] We visually delineated all closed water bodies from each image that were above a minimum area of 0.2 ha. The shoreline of each closed pond was manually traced as a polygon area using a Geographic Information System (GIS). An inventory of closed water bodies from each time period was then analyzed using the GIS. For each study region, the change in the number of closed water bodies and the area of surface water was estimated between each time period and the 1950s period.

2.3. Meteorological Data

[12] For the closest long-term weather station to each study region, we obtained Summary of the Day data set (daily maximum temperature, daily minimum temperature, and daily total water equivalent precipitation, including snow) from the Alaska Climate Research Center in Fairbanks, Alaska. Unfortunately, Alaska has few stations with weather records spanning more than 50 years and therefore some stations used in the study are located hundreds of kilometers from the center of a study region (Figure 1).

[13] For each weather station, we computed mean monthly temperature as

display math

where n is number of days in each month and Max, Min are daily maximum/minimum temperatures. We used monthly mean temperature to estimate annual potential evapotranspiration (PET) using the Thornthwaite's method [Thornthwaite, 1948]. We then estimated water balance at each weather station as precipitation – PET for each year.

3. Results

[14] All study regions in subarctic Alaska had a reduction in the area of closed ponds, ranging from −31 to −4 percent (Table 2). The Arctic Coastal Plain region had negligible change. The number of closed ponds decreased over the 50-year period for all study regions (Table 3). As surface water decreased, smaller multiple ponds were sometimes created (Figure 2), and therefore the loss of closed basin ponds may be a conservative estimate in some areas. Most of the decrease in number of ponds occurred after a climatic regime shift [Mantua et al., 1997; Hare and Mantua, 2000] in the 1970s (Table 3).

Figure 2.

Shrinkage of open water within the Yukon Flats National Wildlife Refuge. Notice that it is likely to increase the number of smaller ponds as the area of surface water steadily decreases from larger ponds. The series are typical examples of the water loss patterns occurred during the 50-year study period.

Table 2. Area Estimates of Closed Pond Surface Water by Study Regiona
Study RegionAerial PhotographyLandsat TM/ETM+ ImageryPercent Change (1950s – Most Recent Estimate)
  • a

    Units are hectares.

Innoko Flats National Wildlife Refuge212615321483−31
Copper River Basin. Wrangell St. Elias National Park130958593−28
Minto Flats State Game Refuge40663763395233193060−25
Yukon Flats National Wildlife Refuge777380976359−18
Stevens Village51325106476847464398−14
Denali National Park175819641681−4
Tetlin National Wildlife Refuge481145664597−4
Arctic Coastal Plain3162312732081
Totals30,48829,729 26,941 −11.6
Table 3. Total Number of Shallow, Closed-Basin Ponds Inventoried by Study Region
Study RegionAerial PhotographyLandsat TM/ETM+ ImageryPercent Change (1950s – Most Recent Estimate)
Innoko Flats National Wildlife Refuge12051053839−30
Copper River Basin101533746−54
Minto Flats State Game Refuge13901203820934891−36
Yukon Flats National Wildlife Refuge137012671228−10
Stevens Village11411218850896933−18
Denali National Park876964834−5
Tetlin National Wildlife Refuge215222931725−20
Arctic Coastal Plain142614481265−11
Totals10,71310,525 8643 −19

[15] Except for Barrow, we found a significant (p < 0.05) linear increase in mean annual temperature and annual total PET from all first-order weather stations (Table 4 and Figure 3). There was no significant trend for total annual precipitation at these stations. Water balance, defined as annual total precipitation – annual total PET (Table 4) had a consistently negative trend for all stations except Bettles. The Bettles station is located in the foothills of the Brooks Range, 260 km away from the Yukon Flats study region (Figure 1) and may not have been representative of that study region.

Figure 3.

Ten-year running means of potential evapotranspiration (PET) for long-term, first-order weather stations. The linear trend since 1970 was positive and significant (P < 0.01) for all stations analyzed.

Table 4. Linear Temporal Trends in Mean Annual Temperature, Annual Precipitation, Annual Potential Evapotranspiration (PET), and Water Balance Based on the Closest Long-Term, First-Order Weather Station to Each Study Regiona
Study RegionsWeather StationMean Annual Temperature, °CTotal Annual Precipitation, mmTotal Annual PET, mmAnnual Water Balance
  • a

    Water balance is annual precipitation – PET.

Arctic Coastal PlainBarrow Airport (1941–2002)0.0010.0440.116−0.4820.0500.0890.7610.0980.016−1.2430.1500.002
Yukon FlatsBettles Airport (1951–2002)0.0460.2590.00011.1890.0350.1900.6070.1310.0090.5820.0070.559
Minto Flats/Stevens Village/Denali National ParkFairbanks Airport (1946–2002)0.0370.2090.0005−0.7070.0180.3400.7160.1700.002−0.7070.1760.340
Tetlin National Wildlife RefugeNorthway Airport (1949–2002)0.0340.2090.00080.4910.01780.3500.379−.0680.064−0.1110.0007.8511
Copper River BasinGulkana Airport (1941–2002)0.0160.0630.0550.1500.0020.7410.3920.0100.015−0.2430.0040.621
Innoko Flats National Wildlife RefugeMcGrath Airport (1939–2002)0.0260.1360.003−0.2510.0020.7430.4660.1500.002−0.7170.0150.354
TalkeetnaTalkeetna Airport (1940–2002)0.0240.1160.007−0.3540.0010.7710.3490.0700.039−0.7030.0050.572

4. Discussion

[16] There are several nonclimatic potential reasons for the long-term trend in pond shrinkage. One source of potential error in our estimates is the larger pixel size (30 m) in our latest period using satellite imagery relative to the 1950s and 1978–1982 aerial photography (3 m pixel size). However, we believe there was a tendency to overestimate pond areas using satellite imagery since water strongly absorbs midinfrared radiation and subpixel water absorption typically leads to an overestimate of “water pixels.” There was a reduction in closed pond area for 6 of the 8 subarctic study regions from 1950s to 1978–1982, based exclusively on aerial photography (Table 2).

[17] If the first period in the time series was an unusually wet year, then the trend might be due to a sampling artifact within the range of natural variability of precipitation. We normalized precipitation of every year corresponding to our study image years as

display math

where Precip is September through August total precipitation for image year, Mean Precip is Mean September through August total precipitation from 1950–2004, and Std. Dev. is Standard Deviation of September through August total precipitation. For most of the periods corresponding to our remotely sensed images, the normalized precipitation was within 1 standard deviation of the 1950–2004 mean (Figure 4).

Figure 4.

Growing season precipitation (May through August) scaled as normal standard deviate by first-order weather stations closest to each study area.

[18] We believe that increased evapotranspiration due to warmer and longer growing seasons is one factor associated with the long-term trend of shrinking pond surface areas. On the basis of analysis of 1988–2002 satellite microwave data [Smith et al., 2004], the growing season (spring thaw to autumn freeze) increased about 5 days decade−1, and within Alaska was highest in our study region of interior Alaska. While our analyses did not identify a significant increase in annual water deficit since 1950, other studies have identified that summer (June–August) water deficits have increased in interior Alaska by 5.5 mm year−1 since the 1960 [Hinzman et al., 2005]. On the basis of tree-ring analysis, drought-stress of white spruce trees in interior Alaska has occurred during the warming period since 1950 [Barber et al., 2000; Lloyd and Fastie, 2002]. There has also been a substantial reduction in the area and number of water bodies on the Kenai Peninsula, Alaska, due to a decrease in water balance since 1950 [Klein et al., 2005]. In contrast, we found no reduction in the area and number of water bodies in coastal Arctic Alaska, although summer water deficits have increased in that region during recent decades [Oechel et al., 2000] [see also Hinzman et al., 2005]. Coastal Arctic Alaska is characterized by a thaw-lake cycle that involves drainage leading to several thousand years of vegetation succession followed by lake development that ultimately results in drainage [Hopkins, 1949; Billings and Peterson, 1980; Hinkel et al., 2003; Bockheim et al., 2004]. Our analysis suggests that the thaw-lake cycle in our study region of Arctic Alaska has been stable over the last 50 years.

[19] Shrinking pond surface areas may become a common feature in discontinuous permafrost regions as a consequence of warming climate and thawing permafrost [Yoshikawa and Hinzman, 2003]. Another mechanism for pond drainage is the expansion of a layer of unfrozen soil (talik) allowing drainage. Smith et al. [2005] used satellite imagery from the 1970s and 1997–2004 to document the decline in lake abundance and surface area for thousands of lakes in Siberia. Their analysis showed an increase of total lake area and numbers in a continuous permafrost region, and a contrasting decrease in regions of discontinuous, sporadic, and isolated permafrost. Yoshikawa and Hinzman [2003] used remotely sensed imagery to document a decrease in surface area of thermokarst ponds in western Alaska. They used ground penetrating radar to detect talik that would allow subsurface drainage below the shrinking ponds. Smith et al. [2005] observed permanently drained lakes commonly alongside undrained neighboring lakes, suggesting a spatially patchy process related to permafrost distribution rather than a direct climatic mechanism such as increased evaporation. We also found a heterogeneous pattern of shrinking ponds within some study regions. For example, in the Yukon Flats study region, there was a slight increase in closed pond surface area at a higher elevation area, while lower elevations had a decrease of over 18 percent. Even within the extent of a single aerial photograph, there was a heterogeneous pattern of shrinking ponds, with some ponds remaining stable since the 1950s while neighboring ponds decreased substantially in surface area (Figures 5a and 5b). This pattern may be related to the distribution of discontinuous permafrost and subsurface drainage related to warming permafrost [Yoshikawa and Hinzman, 2003].

Figure 5.

Heterogeneous pattern of shrinking ponds. For example, ponds A, B, and C exhibit little change in surface water area, while ponds D, E, and F shrink substantially since the 1950s.

Figure 5.


[20] Wildfire in the Alaska boreal forest can accelerate eventual warming of permafrost by removing much of the insulating duff layer and increasing soil thermal conductivity and ground heat flux [Yoshikawa et al., 2003; Chambers and Chapin, 2002]. In a warming climate, the loss of a frozen soil aquiclude may promote long-term soil drying [Swanson, 1996], especially with the increase in transpiration as prefire black spruce is replaced by early succession broadleaf species [Betts et al., 1999]. Thus wildfire legacies may be another factor responsible for the observed pattern of shrinking ponds. On the basis of known fires recorded by the Alaska Fire Service, some of the study regions had over 25 percent of their areas covered by known wildfires since the 1950s (Table 5).

Table 5. Percent of Study Regions Burned by 1950–2000 Fires Mapped by the Alaska Fire Service
Study RegionArea Burned, %
Innoko Flats National Wildlife Refuge20.0
Copper River Basin. Wrangell St. Elias National Park0.2
Minto Flats State Game Refuge32.4
Yukon Flats National Wildlife Refuge18.3
Stevens Village37.5
Denali National Park13.5
Tetlin National Wildlife Refuge10.9
Arctic Coastal Plain0.0

[21] Since wildfire, permafrost thawing, and evapotranspiration may have increased in our study regions since the 1950s, it is difficult to determine the dominant driving mechanism behind the observed pattern of shrinking ponds. All three mechanisms have been documented in boreal Alaska. For example, in a study of 11 wildfire burns in interior Alaska, Yoshikawa et al. [2003] found that all burns older than 10 years had lower soil moisture content relative to adjacent unburned control areas. The long-term effect of wildfire on decreased soil moisture may be due to both increased transpiration and thawing of permafrost. In the permafrost-dominated Seward Peninsula, Yoshikawa and Hinzman [2003] found a reduction in thermokarst ponds over ice wedge terrain exceeding 80 percent based on 1950 to 1981 aerial photography. They concluded the major mechanism for shrinking ponds was formation of taliks allowing internal drainage through the permafrost. In the permafrost-free Kenai Peninsula of Alaska, Klein et al. [2005] found a reduction in pond area exceeding 70 percent for two subregions based on 1950 to 1996 aerial photography. The authors concluded the major mechanism for shrinking ponds was decreased water balance in a warming and drying climate.

[22] The reduction in the surface area of closed-basin lakes that we have detected in this study may be the initial signal of more widespread changes that are occurring in low lying areas of interior Alaska. In particular, this signal may be indicative of widespread lowering of the water table throughout low-lying landscapes in interior Alaska. A lowering of the water table would affect the structure and function of wetlands adjacent to lakes and would likely drive vegetation dynamics from the conversion of wetlands towards upland vegetation. These changes have the potential to affect two classes of ecosystem services from wetlands in Alaska: (1) climate regulation services associated with the exchange of radiatively active gases with the atmosphere and (2) provisioning services affecting natural resources used by people. Climate regulation services are likely to be affected in a complex manner as a lowering of the water table can enhance the release of carbon dioxide by exposing soil carbon to aerobic decomposition and the delivery of dissolved organic carbon to coastal ecosystems, but would likely decrease the emissions of methane to the atmosphere [McGuire et al., 2006]. However, a lowering of the water table could have profound consequences for provisioning services and the management of natural resources on National Wildlife Refuges in Alaska, which cover over 77 million acres and comprise 81% of the National Wildlife Refuge System. These refuges provide breeding habitat for millions of waterfowl and shorebirds that winter in more southerly regions of North America. Wetland areas have also been traditionally important in the subsistence lifestyles of native peoples in interior Alaska, and changes in the structure and function of wetlands has the potential to affect the sustainability of subsistence lifestyles. Because the signal of reduced surface area of closed-basin lakes in this study may have implications for ecosystem services both within Alaska and throughout North America, there is need to extend the analyses presented in this study with analyses that more fully assess the historical trends, variability, geographic scope, and mechanisms of lake drying in interior Alaska.


[23] This project was funded through support from the NASA Land Cover/Land Use Change Program (NAF-11142) and the Bonanza Creek LTER (Long-Term Ecological Research) program (funded jointly by NSF grant DEB-0423442 and USDA Forest Service, Pacific Northwest Research Station grant PNW01-JV11261952-231).